Re: [RFC PATCH] cgroup/cpuset: fix a memory binding failure for cgroup v2

[Tejun Heo <tj@kernel.org>]

cc'ing Waiman and copying the whole body.

Waiman, can you please take a look?

Thanks.

On Tue, Apr 19, 2022 at 10:09:58AM +0800, Feng Tang wrote:
> We got report that setting cpuset.mems failed when the nodemask
> contains a newly onlined memory node (not enumerated during boot)
> for cgroup v2, while the binding succeeded for cgroup v1.
> 
> The root cause is, for cgroup v2, when a new memory node is onlined,
> top_cpuset's 'mem_allowed' is not updated with the new nodemask of
> memory nodes, and the following setting memory nodemask will fail,
> if the nodemask contains a new node.
> 
> Fix it by updating top_cpuset.mems_allowed right after the
> new memory node is onlined, just like v1.
> 
> Signed-off-by: Feng Tang <feng.tang@intel.com>
> ---
> Very likely I missed some details here, but it looks strange that
> the top_cpuset.mem_allowed is not updatd even after we onlined
> several memory nodes after boot.
> 
>  kernel/cgroup/cpuset.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index 9390bfd9f1cd..b97caaf16374 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -3314,8 +3314,7 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
>  	/* synchronize mems_allowed to N_MEMORY */
>  	if (mems_updated) {
>  		spin_lock_irq(&callback_lock);
> -		if (!on_dfl)
> -			top_cpuset.mems_allowed = new_mems;
> +		top_cpuset.mems_allowed = new_mems;
>  		top_cpuset.effective_mems = new_mems;
>  		spin_unlock_irq(&callback_lock);
>  		update_tasks_nodemask(&top_cpuset);
> -- 
> 2.27.0
> 

-- 
tejun